Shelton, PeirisGadhi, Adel2025-08-182025-08https://hdl.handle.net/20.500.14154/76172This thesis investigates the use of machine learning and hybrid models to forecast time series data such as climate patterns, oil prices, Australian beer production, and sunspot activity. It examines traditional models like ARIMA and GARCH, as well as machine learning methods such as SVR, LSTM, RF, and DT, which better capture non-linear and complex relationships. The study also evaluates hybrid models like ARIMA-ANN and GARMA-LSTM, which consistently demonstrate superior forecasting accuracy across various datasets. The GARMA-LSTM model, in particular, proves effective for long-term forecasting, especially with sunspot and beer production data. Finally, the thesis applies an advanced deep learning system, WGAN-GP, to financial and climate data, showing that modern methods can move beyond classical assumptions and better capture complex, high-order dynamics.212enARIMAARFIMAGARMAWGANANNLSTMDeep LearningHybrid ModelsMachines LearningTime SeriesVolatilityStatisticsThesis